A decensored 1.5B model with strong instruction-following and suppressed refusal behavior. Made by RACER IS OP.

tools

ollama run R4C3R/qwen2.5-1.5b-heretic

curl http://localhost:11434/api/chat \
  -d '{
    "model": "R4C3R/qwen2.5-1.5b-heretic",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='R4C3R/qwen2.5-1.5b-heretic',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'R4C3R/qwen2.5-1.5b-heretic',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 2 months ago

2 months ago

a2652f5a1e60 · 986MB

model

archqwen2

parameters1.54B

quantizationQ4_K_M

986MB

template

{{- if .Suffix }}<|fim_prefix|>{{ .Prompt }}<|fim_suffix|>{{ .Suffix }}<|fim_middle|> {{- else if .M

1.6kB

Readme

🏴 Qwen2.5-1.5B-Heretic

Heretic Series by RACER IS OP

Qwen2.5-1.5B-Heretic is a decensored variant of Qwen’s compact 1.5B instruct model. Knowledge and instruction-following are preserved while refusal behavior is removed via targeted ablation.

Who this is for — developers who want a light 1.5B instruct model without guardrails. Great for local agents, summarization, and general-purpose chat on modest hardware.

Quickstart

ollama run R4C3R/qwen2.5-1.5b-heretic

VRAM Guide (choose your quant)

Quant	VRAM	When to use
`q4_k_m`	~0.9 GB	Best balance — runs on any system
`q5_k_m`	~1.1 GB	Higher quality, still lightweight
`q8_0`	~1.6 GB	Maximum quality, CPU-friendly

On a tight system? Start with q4_k_m. Have room to spare? q8_0 gives you the full model quality.

Links

sdad.pro · Heretic

License: qwen-research

Made with love by RACER IS OP — follow for more uncensored models in the Heretic Series.

# 🏴 Qwen2.5-1.5B-Heretic

> **Heretic Series** by RACER IS OP

---

Qwen2.5-1.5B-Heretic is a decensored variant of Qwen's compact 1.5B instruct model. Knowledge and instruction-following are preserved while refusal behavior is removed via targeted ablation.

**Who this is for** — developers who want a light 1.5B instruct model without guardrails. Great for local agents, summarization, and general-purpose chat on modest hardware.

---

### Quickstart

```bash
ollama run R4C3R/qwen2.5-1.5b-heretic
```

### VRAM Guide (choose your quant)

| Quant | VRAM | When to use |
|-------|------|-------------|
| `q4_k_m` | ~0.9 GB | Best balance — runs on any system |
| `q5_k_m` | ~1.1 GB | Higher quality, still lightweight |
| `q8_0` | ~1.6 GB | Maximum quality, CPU-friendly |

> On a tight system? Start with `q4_k_m`. Have room to spare? `q8_0` gives you the full model quality.

### Links

[sdad.pro](https://sdad.pro) · [Heretic](https://github.com/p-e-w/heretic)

**License:** [qwen-research](https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct/blob/main/LICENSE)

---

*Made with love by RACER IS OP — follow for more uncensored models in the Heretic Series.*

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)